Spectral Data Classification Using Odds Ratios

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for classifying biological samples using spectrographic data, such as mass spectrometry and Raman spectrometry, often prioritize the presence of peaks over the absence and fail to adequately account for variability within biological clusters, leading to suboptimal classification accuracy.

Innovation Solution

A method that calculates the probability of peak occurrence and uses odds ratios to classify samples by comparing the spectral data of unknown samples to established clusters, considering both the presence and absence of peaks, and incorporating a weighing factor for peak amplitude, while allowing for self-learning and dynamic adaptation of reference spectra.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional classification methods compare individual spectra to reference spectra and classify based on the most similar spectrum, then the classification process is simple and fast, but the accuracy is suboptimal because more weight is given to the presence of peaks rather than the absence of peaks and variability within biological clusters is not adequately compensated

Engineering Contradiction:
Improveclassification accuracyVSAvoidcomplexity of classification algorithm
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent transforms the classification approach by changing the parameters used for comparison: instead of directly comparing spectral intensities, it converts peak presence/absence into probability values and then into odds ratios. This parameter transformation allows the system to weigh both presence and absence of peaks appropriately while compensating for biological variability, thereby improving classification accuracy without requiring excessively complex algorithms

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent performs preliminary calculations of probability of peak occurrence and odds ratios for each cluster before actual classification. By pre-computing these statistical parameters from reference spectra and storing them for quick retrieval, the system prepares the necessary classification criteria in advance, enabling accurate real-time classification without complex on-the-fly computations

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If the system uses probability calculations and odds ratios to account for peak absence and variability, then classification accuracy improves, but the computational complexity increases

Engineering Contradiction:
Improveclassification accuracyVSAvoidcomputational power required
Core Design Contradiction:
Measurement precisionVSPower

Solution Approach 1:

The system performs computationally intensive probability and odds ratio calculations in advance during the reference spectrum preparation phase. These pre-computed statistical parameters are stored and reused during actual classification, shifting the computational burden to an offline setup phase rather than requiring high computational power during real-time sample analysis

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent focuses computational effort on the most discriminative peaks by calculating probabilities and odds ratios only for peaks that are relevant to cluster differentiation. Rather than computing exhaustive comparisons for all possible spectral features, the method selectively processes only the necessary statistical parameters, reducing overall computational requirements while maintaining accuracy

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP2622355B1New classification method for spectral data
Publication Date: 2015.03.04 NEDERLANDSE ORG VOOR TOEGEPAST NATUURWETENSCHAPPELIJK ONDERZOEK TNO
  • EP2622355B1 patent drawingFigure 1
  • EP2622355B1 patent drawingFigure 2
  • EP2622355B1 patent drawing

AI summary

The present invention relates to a new method for classification of spectral data comprising: a. analyzing at least two samples belonging to at least one cluster through recording of a spectrum; b. for each spectrum determining the peaks and the spectral value at which they occur; c. calculating the probability (p) of occurrence for each peak for every cluster and from this the odds ratio p/(1-p) for each peak; d. preparing a spectrum from a sample to be classified with the same technique as in step a); e. determine the peaks in the spectrum obtained in step d); f. calculate the likelihood of identity for each cluster by multiplying per cluster the odds ratio in said cluster for each peak found in the spectrum of step d); and g. assign the sample tested in step d) to the cluster that provides the largest number as a result of step f). The invention also comprises a system for performing such a method and the use of such a system for classification of spectral data.